GWU NLP at SemEval-2019 Task 7: Hybrid Pipeline for Rumour Veracity and Stance Classification on Social Media

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 1115–1119 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 1115 GWU NLP at SemEval-2019 Task 7: Hybrid Pipeline for Rumour Veracity and Stance Classification on Social Media Sardar Hamidian and Mona Diab Department of Computer Science The George Washington University Washington DC, USA {sardar, mtdiab} @gwu.edu Abstract Social media plays a crucial role as the main re- source news for information seekers online. How- ever, the unmoderated feature of social media plat- forms lead to the emergence and spread of un- trustworthy contents which harm individuals or even societies. Most of the current automated approaches for automatically determining the ve- racity of a rumor are not generalizable for novel emerging topics. This paper describes our hybrid system comprising rules and a machine learning model which makes use of replied tweets to iden- tify the veracity of the source tweet. The proposed system in this paper achieved 0.435 F-Macro in stance classification, and 0.262 F-macro and 0.801 RMSE in rumor verification tasks in Task7 of Se- 1mEval 2019. Introduction The number of users who rely on social me- dia to seek daily news and information rises daily, but not all information online is trustworthy. The unmoderated feature of social media makes the emergence and diffusion of misinformation even more intense. Consequently, the propagation of misinformation online could harm an individual or even a society. Most of the current approaches on verifying credibility perform well for the un- fold topics which are already verified by a trust- worthy resource (Qazvinian et al., 2011; Hamid- ian and Diab, 2015). However, the performance suffers when it comes to real life application for dealing with the emerging rumors which are pri- orly unknown. Identifying the emerging rumor and veracity of the rumor by relying on previ- ous observations is a challenging task as the new emerging rumor could be entirely new regarding the event, propagation pattern, and also the prove- nance. Despite these challenges, many researchers have been studying the generalizable metrics that could be aggregated from the source, replied posts, or network information (Vosoughi et al., 2018b; Kochkina et al., 2018; Grinberg et al., 2019). Our first mission in this paper is to automatically de- termine the veracity of rumors as part of the Se- mEval task. SemEval is an ongoing shared task for evaluations of computational sentiment anal- ysis systems. Task 7 (RumourEval19) (Gorrell et al., 2019) is one of the twelve tasks, consist- ing of two subtasks. Task A is about stance orien- tation of people as supporting, denying, querying or commenting (SDQC) in a rumor discourse and Task B is about the verification of a given rumor. We propose a hybrid model with rules and a neural network machine learning scheme for both tasks. For task A we rely on the text content of the post, and its parent. In Task B not only do we aggregate contextual information of the source of the rumor but also …

Content truncated to 500 words